Does yolov5 has the data augmentation of flip horizontally ?

[wwdok]

❔Question

I see some basic data augmentations in train.py, but i don't see flip horizontally and mosaic and so on, so does yolov5 support other more data augmentations ?

[github-actions]

Hello @wwdok, thank you for your interest in our work! Please visit our Custom Training Tutorial to get started, and see our Jupyter Notebook , Docker Image, and Google Cloud Quickstart Guide for example environments.

If this is a bug report, please provide screenshots and minimum viable code to reproduce your issue, otherwise we can not help you.

If this is a custom model or data training question, please note Ultralytics does not provide free personal support. As a leader in vision ML and AI, we do offer professional consulting, from simple expert advice up to delivery of fully customized, end-to-end production solutions for our clients, such as:

• Cloud-based AI systems operating on hundreds of HD video streams in realtime.
• Edge AI integrated into custom iOS and Android apps for realtime 30 FPS video inference.
• Custom data training, hyperparameter evolution, and model exportation to any destination.

For more information please visit https://www.ultralytics.com.

[glenn-jocher]

Augmentation hyperparameters are located here:

yolov5/data/hyp.scratch.yaml

Lines 1 to 33 in 83deec1

	# Hyperparameters for COCO training from scratch
	# python train.py --batch 40 --cfg yolov5m.yaml --weights '' --data coco.yaml --img 640 --epochs 300
	# See tutorials for hyperparameter evolution https://github.com/ultralytics/yolov5#tutorials
	lr0: 0.01 # initial learning rate (SGD=1E-2, Adam=1E-3)
	lrf: 0.2 # final OneCycleLR learning rate (lr0 * lrf)
	momentum: 0.937 # SGD momentum/Adam beta1
	weight_decay: 0.0005 # optimizer weight decay 5e-4
	warmup_epochs: 3.0 # warmup epochs (fractions ok)
	warmup_momentum: 0.8 # warmup initial momentum
	warmup_bias_lr: 0.1 # warmup initial bias lr
	box: 0.05 # box loss gain
	cls: 0.5 # cls loss gain
	cls_pw: 1.0 # cls BCELoss positive_weight
	obj: 1.0 # obj loss gain (scale with pixels)
	obj_pw: 1.0 # obj BCELoss positive_weight
	iou_t: 0.20 # IoU training threshold
	anchor_t: 4.0 # anchor-multiple threshold
	# anchors: 0 # anchors per output grid (0 to ignore)
	fl_gamma: 0.0 # focal loss gamma (efficientDet default gamma=1.5)
	hsv_h: 0.015 # image HSV-Hue augmentation (fraction)
	hsv_s: 0.7 # image HSV-Saturation augmentation (fraction)
	hsv_v: 0.4 # image HSV-Value augmentation (fraction)
	degrees: 0.0 # image rotation (+/- deg)
	translate: 0.1 # image translation (+/- fraction)
	scale: 0.5 # image scale (+/- gain)
	shear: 0.0 # image shear (+/- deg)
	perspective: 0.0 # image perspective (+/- fraction), range 0-0.001
	flipud: 0.0 # image flip up-down (probability)
	fliplr: 0.5 # image flip left-right (probability)
	mosaic: 1.0 # image mosaic (probability)
	mixup: 0.0 # image mixup (probability)

[wwdok]

@glenn-jocher Sorry, i find my repo is out of date, the newest repo already has fliplr hypeparameter ! Thanks !

[glenn-jocher]

@wwdok ah, you should git pull often, as repo changes nightly.

[wwdok]

@wwdok ah, you should git pull often, as repo changes nightly.

Hi, bro, I find in the train.py of 3 months ago, there is mixed_precision, but now i find the latest train.py does not include mixed_precision, why was it cancelled ?

[glenn-jocher]

@wwdok mixed precision is integrated by default now.

[wwdok]

@glenn-jocher Great ! 😀

[LogicNg]

So in every epochs the data will be augmented according to the probability?

[glenn-jocher]

@LogicNg yes every mosaic.

[julian-douglas]

I have a question about the augmentations. I don't understand how the HSV augmentations are generated. For example, we have:
hsv_s: 0.7 # image HSV-Saturation augmentation (fraction)

What does this do? Is this changing the saturation to 70% of the original? Or is it ± 70%, so 30% and 170%? Or is it a 0.7 probability that the saturation will change? If so, how much will it change by? Will it both increase and decrease? Will it affect every image?

Also, for the image rotations, if I put 90, will that be clockwise or anticlockwise?

Thank you

[glenn-jocher]

@julian-douglas augmentations are defined in your hyperparameter file and implemented in datasets.py. Albumentations may be added additionally, see Albumentations PR.

[RainbowSun11Q2H]

Thanks for your code.

Does the "degree" in hyperparameter files mean rotation augmentation? And what is the range of "degree"?

Thanks.

[glenn-jocher]

@RainbowSun11Q2H 👋 Hello! Thanks for asking about image augmentation. degree limits are +/- 180

YOLOv5 🚀 applies online imagespace and colorspace augmentations in the trainloader (but not the val_loader) to present a new and unique augmented Mosaic (original image + 3 random images) each time an image is loaded for training. Images are never presented twice in the same way.

Augmentation Hyperparameters

The hyperparameters used to define these augmentations are in your hyperparameter file (default data/hyp.scratch.yaml) defined when training:

python train.py --hyp hyp.scratch-low.yaml

yolov5/data/hyps/hyp.scratch-low.yaml

Lines 6 to 34 in b94b59e

	lr0: 0.01 # initial learning rate (SGD=1E-2, Adam=1E-3)
	lrf: 0.01 # final OneCycleLR learning rate (lr0 * lrf)
	momentum: 0.937 # SGD momentum/Adam beta1
	weight_decay: 0.0005 # optimizer weight decay 5e-4
	warmup_epochs: 3.0 # warmup epochs (fractions ok)
	warmup_momentum: 0.8 # warmup initial momentum
	warmup_bias_lr: 0.1 # warmup initial bias lr
	box: 0.05 # box loss gain
	cls: 0.5 # cls loss gain
	cls_pw: 1.0 # cls BCELoss positive_weight
	obj: 1.0 # obj loss gain (scale with pixels)
	obj_pw: 1.0 # obj BCELoss positive_weight
	iou_t: 0.20 # IoU training threshold
	anchor_t: 4.0 # anchor-multiple threshold
	# anchors: 3 # anchors per output layer (0 to ignore)
	fl_gamma: 0.0 # focal loss gamma (efficientDet default gamma=1.5)
	hsv_h: 0.015 # image HSV-Hue augmentation (fraction)
	hsv_s: 0.7 # image HSV-Saturation augmentation (fraction)
	hsv_v: 0.4 # image HSV-Value augmentation (fraction)
	degrees: 0.0 # image rotation (+/- deg)
	translate: 0.1 # image translation (+/- fraction)
	scale: 0.5 # image scale (+/- gain)
	shear: 0.0 # image shear (+/- deg)
	perspective: 0.0 # image perspective (+/- fraction), range 0-0.001
	flipud: 0.0 # image flip up-down (probability)
	fliplr: 0.5 # image flip left-right (probability)
	mosaic: 1.0 # image mosaic (probability)
	mixup: 0.0 # image mixup (probability)
	copy_paste: 0.0 # segment copy-paste (probability)

Augmentation Previews

You can view the effect of your augmentation policy in your train_batch*.jpg images once training starts. These images will be in your train logging directory, typically yolov5/runs/train/exp:

train_batch0.jpg shows train batch 0 mosaics and labels:

YOLOv5 Albumentations Integration

YOLOv5 🚀 is now fully integrated with Albumentations, a popular open-source image augmentation package. Now you can train the world's best Vision AI models even better with custom Albumentations 😃!

PR #3882 implements this integration, which will automatically apply Albumentations transforms during YOLOv5 training if albumentations>=1.0.3 is installed in your environment. See #3882 for full details.

Example train_batch0.jpg on COCO128 dataset with Blur, MedianBlur and ToGray. See the YOLOv5 Notebooks to reproduce:

Good luck 🍀 and let us know if you have any other questions!

[Bara-Elba]

Can I indicate how many images exactly to be generated.
I have a small dataset and I want to augment it *10, I searched in the code and didn't find where to do it.
is there a possibility to do so ?

[glenn-jocher]

@Bara-Elba 👋 Hello! Thanks for asking about image augmentation. YOLOv5 🚀 applies online imagespace and colorspace augmentations in the trainloader (but not the val_loader) to present a new and unique augmented Mosaic (original image + 3 random images) each time an image is loaded for training. Images are never presented twice in the same way.

Augmentation Hyperparameters

The hyperparameters used to define these augmentations are in your hyperparameter file (default data/hyp.scratch.yaml) defined when training:

python train.py --hyp hyp.scratch-low.yaml

yolov5/data/hyps/hyp.scratch-low.yaml

Lines 6 to 34 in b94b59e

	lr0: 0.01 # initial learning rate (SGD=1E-2, Adam=1E-3)
	lrf: 0.01 # final OneCycleLR learning rate (lr0 * lrf)
	momentum: 0.937 # SGD momentum/Adam beta1
	weight_decay: 0.0005 # optimizer weight decay 5e-4
	warmup_epochs: 3.0 # warmup epochs (fractions ok)
	warmup_momentum: 0.8 # warmup initial momentum
	warmup_bias_lr: 0.1 # warmup initial bias lr
	box: 0.05 # box loss gain
	cls: 0.5 # cls loss gain
	cls_pw: 1.0 # cls BCELoss positive_weight
	obj: 1.0 # obj loss gain (scale with pixels)
	obj_pw: 1.0 # obj BCELoss positive_weight
	iou_t: 0.20 # IoU training threshold
	anchor_t: 4.0 # anchor-multiple threshold
	# anchors: 3 # anchors per output layer (0 to ignore)
	fl_gamma: 0.0 # focal loss gamma (efficientDet default gamma=1.5)
	hsv_h: 0.015 # image HSV-Hue augmentation (fraction)
	hsv_s: 0.7 # image HSV-Saturation augmentation (fraction)
	hsv_v: 0.4 # image HSV-Value augmentation (fraction)
	degrees: 0.0 # image rotation (+/- deg)
	translate: 0.1 # image translation (+/- fraction)
	scale: 0.5 # image scale (+/- gain)
	shear: 0.0 # image shear (+/- deg)
	perspective: 0.0 # image perspective (+/- fraction), range 0-0.001
	flipud: 0.0 # image flip up-down (probability)
	fliplr: 0.5 # image flip left-right (probability)
	mosaic: 1.0 # image mosaic (probability)
	mixup: 0.0 # image mixup (probability)
	copy_paste: 0.0 # segment copy-paste (probability)

Augmentation Previews

You can view the effect of your augmentation policy in your train_batch*.jpg images once training starts. These images will be in your train logging directory, typically yolov5/runs/train/exp:

train_batch0.jpg shows train batch 0 mosaics and labels:

YOLOv5 Albumentations Integration

YOLOv5 🚀 is now fully integrated with Albumentations, a popular open-source image augmentation package. Now you can train the world's best Vision AI models even better with custom Albumentations 😃!

PR #3882 implements this integration, which will automatically apply Albumentations transforms during YOLOv5 training if albumentations>=1.0.3 is installed in your environment. See #3882 for full details.

Example train_batch0.jpg on COCO128 dataset with Blur, MedianBlur and ToGray. See the YOLOv5 Notebooks to reproduce:

Good luck 🍀 and let us know if you have any other questions!

[Titya79]

Augmentation hyperparameters are located here:

yolov5/data/hyp.scratch.yaml

Lines 1 to 33 in 83deec1

Hyperparameters for COCO training from scratch

python train.py --batch 40 --cfg yolov5m.yaml --weights '' --data coco.yaml --img 640 --epochs 300

See tutorials for hyperparameter evolution https://github.com/ultralytics/yolov5#tutorials

lr0: 0.01 # initial learning rate (SGD=1E-2, Adam=1E-3)
lrf: 0.2 # final OneCycleLR learning rate (lr0 * lrf)
momentum: 0.937 # SGD momentum/Adam beta1
weight_decay: 0.0005 # optimizer weight decay 5e-4
warmup_epochs: 3.0 # warmup epochs (fractions ok)
warmup_momentum: 0.8 # warmup initial momentum
warmup_bias_lr: 0.1 # warmup initial bias lr
box: 0.05 # box loss gain
cls: 0.5 # cls loss gain
cls_pw: 1.0 # cls BCELoss positive_weight
obj: 1.0 # obj loss gain (scale with pixels)
obj_pw: 1.0 # obj BCELoss positive_weight
iou_t: 0.20 # IoU training threshold
anchor_t: 4.0 # anchor-multiple threshold

anchors: 0 # anchors per output grid (0 to ignore)

fl_gamma: 0.0 # focal loss gamma (efficientDet default gamma=1.5)
hsv_h: 0.015 # image HSV-Hue augmentation (fraction)
hsv_s: 0.7 # image HSV-Saturation augmentation (fraction)
hsv_v: 0.4 # image HSV-Value augmentation (fraction)
degrees: 0.0 # image rotation (+/- deg)
translate: 0.1 # image translation (+/- fraction)
scale: 0.5 # image scale (+/- gain)
shear: 0.0 # image shear (+/- deg)
perspective: 0.0 # image perspective (+/- fraction), range 0-0.001
flipud: 0.0 # image flip up-down (probability)
fliplr: 0.5 # image flip left-right (probability)
mosaic: 1.0 # image mosaic (probability)
mixup: 0.0 # image mixup (probability)

I try many time on this, but the result of args.yaml shown not correct with the modification.

1. Modify the data.yaml file for calling train.
   model = YOLO("yolov5nu.pt") # Load a pretrained model (recommended for training)
   yaml = "data_aug.yaml"
   results = model.train(data=yaml, epochs=1, imgsz=640)

data_aug.yml file:

Ultralytics 🚀 AGPL-3.0 License - https://ultralytics.com/license

Train/val/test directory

path: C:/srt/raw_data/datasets_resized2 # dataset root dir
train: images/train # train images
val: images/val # val images
test: # test images (optional)

Classes

names:
0: Traffic Sign

this parameters are all zero since we want to use albumentation framework

fl_gamma: 0.0 # focal loss gamma (efficientDet default gamma=1.5)
hsv_h: 0 # image HSV-Hue augmentation (fraction)
hsv_s: 0 # image HSV-Saturation augmentation (fraction)
hsv_v: 0 # image HSV-Value augmentation (fraction)
degrees: 0.0 # image rotation (+/- deg)
translate: 0 # image translation (+/- fraction)
scale: 0 # image scale (+/- gain)
shear: 0 # image shear (+/- deg)
perspective: 0.0 # image perspective (+/- fraction), range 0-0.001
flipud: 0.0 # image flip up-down (probability)
fliplr: 0.0 # image flip left-right (probability)
mosaic: 0.0 # image mosaic (probability)
mixup: 0.0 # image mixup (probability)
copy_paste: 0.0 # segment copy-paste (probability)

2. Modify the base [yolov5/data/hyps/hyp.scratch-low.yaml] . and calling the train:
   model = YOLO("yolov5nu.pt")
   yaml = "data.yaml" (no augment set)
   results = model.train(data=yaml, epochs=1, imgsz=640)

Could u tell me the correct way to modify the parameters in yaml file? instead of set in the train code. Since if I set in the train, it works (results = model.train(data=yaml, epochs=1, imgsz=640, fliplr=1)
Is the modify for yolo5 and yolo8 the same?

[pderrenger]

For YOLOv5 augmentation parameter modifications, create a custom hyp.yaml file (copy from hyp.scratch.yaml) with your desired values, then train with --hyp your_hyp.yaml. YOLOv5 and YOLOv8 handle hyperparameters differently - for YOLOv5 use the hyp.yaml file, while YOLOv8 supports direct argument passing in train(). Ensure you're using the latest YOLOv5 code with git pull.

[Titya79]

Thank you so much...

[pderrenger]

You're welcome! We're glad you find YOLOv5 useful. All credit goes to our amazing open-source community and contributors. Let us know if you have any specific questions as you explore the project further! 🚀